Reconstruction of Gamma-ray interaction in Thick, Monolithic Scintillators via Neural Network-Based Techniques

The reconstruction of the position of interaction in thick, monolithic scintillator crystals is a fundamental challenge in various fields, such as nuclear physics, astrophysics and nuclear medicine, e.g. PGI (Prompt Gamma Imaging) and BNCT (Boron Neutron Capture Therapy). Monolithic crystals have the main advantage of achieving high energy resolution, which is crucial for accurate spectroscopy measurements. However, when the scintillator is thick for good efficiency, large light spread, Compton events, internal reflections and optical grease coupling, can limit the spatial resolution, leading to errors in position reconstruction. Artificial intelligence (AI) techniques, such as Artificial Neural Networks (ANNs), have emerged as a promising solution to address these issues. They have the capability to learn from data and improve their performances over time, making them ideal candidates for achieving good precision and accuracy in the reconstruction. In this case of study, we report the experimental characterization of a channel-edge pinhole collimator specifically designed for BNCT-SPECT applications and the imaging experimental results obtained by combining the collimator with a detection module based on a squared 5 cm×5 cm×2 cm LaBr3(Ce+Sr) monolithic scintillator crystal. The overall system, coupled to a fully connected ANN, proves the possibility to track and resolve sources of radiation with good spatial resolution, of about 3.25 mm.